基于NPP/VIIRS夜光遥感数据的淮安市夜间PM_(2.5)浓度估算研究  

Estimating nighttime PM_(2.5) concentrations in Huai’an based on NPP/VIIRS nighttime light data

作　　者：陈惠娟 徐永明[1] 莫亚萍 张悦[2] 杨子毅 CHEN Huijuan;XU Yongming;MO Yaping;ZHANG Yue;YANG Ziyi(Nanjing University of Information Science&Technology,Nanjing 210044;Environmental Monitoring Center of Jiangsu Province,Nanjing 210036;Huai’an Environmental Monitoring Center of Jiangsu Province,Huai’an 223001)

机构地区：[1]南京信息工程大学,南京210044 [2]江苏省环境监测中心,南京210036 [3]江苏省淮安环境监测中心,淮安223001

出　　处：《环境科学学报》2022年第3期342-351,共10页Acta Scientiae Circumstantiae

基　　金：江苏省环境监测科研基金项目(No.1903);国家自然科学基金面上项目(No.41871028);江苏省青蓝工程(No.R2019Q03)。

摘　　要：PM_(2.5)是大气的重要污染物,掌握其空间分布对于大气污染防控具有重要意义.目前,PM_(2.5)遥感监测主要围绕卫星反演的日间AOD数据开展,无法反映夜间大气污染的空间格局.以2019年9—12月NPP/VIIRS夜间灯光影像和空气质量站点PM_(2.5)观测数据对江苏省淮安市夜间PM_(2.5)浓度进行估算研究.基于辐射传输方程分析夜间灯光辐射与PM_(2.5)浓度之间的关系,在此基础上综合考虑灯光辐射直接衰减和散射补偿确定了计算夜间PM_(2.5)浓度的空间自变量,运用多元线性回归模型(MLR)、随机森林(RF)、Cubist、极端梯度提升树(XGBoost)、神经网络(NNet)、支持向量机(SVM)及最近邻法(KNN)算法构建夜间PM_(2.5)浓度遥感估算模型.结果表明,多元线性归回模型精度明显低于各个机器学习模型,所有模型中SVM模型精度最高,决定系数R^(2)为0.77,平均绝对误差MAE为20.83μg·m^(-3),均方根误差RMSE为32.05μg·m^(-3).基于建立的SVM模型估算了淮安市夜间PM_(2.5)浓度,并对其空间分布特征进行了分析.本研究探索了利用夜间灯光遥感数据估算夜间PM_(2.5)浓度的方法,为夜间大气环境监测与管理提供了参考.PM_(2.5)is an important pollutant in the atmosphere,and detailed knowledge of its spatial distribution is essential for air pollution prevention and control.Most previous studies have focused on estimating PM_(2.5)concentrations from satellite derived daytime AOD data,which cannot effectively depict nighttime atmospheric environment.This paper aims to estimate the nighttime PM_(2.5)concentrations in Huai'an City,Jiangsu Province,using NPP/VIIRS nighttime light data and station observed PM_(2.5)data during September-December 2019.The relationship between nighttime light radiation and PM_(2.5)concentration was first explored based on radiative transfer equation.Taking into account direct attenuation of light radiation and scattering compensation,the spatial independent variables for nighttime PM_(2.5)estimation were determined.Multiple Linear Regression(MLR),Random Forest(RF),Cubist,Extreme Gradient Boosting Tree(XGBoost),Neural Network(NNet),Support Vector Machine(SVM)and Nearest Neighbor Method(KNN)were used to develop remote sensing models for estimating nighttime PM_(2.5)concentrations.The results showed that MLR model had a significantly lower accuracy than machine learning models,and SVM model outperformed other models,with a coefficient of determination(R^(2))of 0.77,a mean absolute error(MAE)of 20.83μg·m^(-3)and a root mean square error(RMSE)of 32.05μg·m^(-3).Huai'an nighttime PM_(2.5)concentration was mapped based on the developed SVM model,and its spatial distribution characteristics were analyzed.This paper explores a method for estimating nighttime PM_(2.5)concentrations with nighttime light remote sensing data,and provides references for the monitoring and management of the nighttime atmospheric environment.

关 键 词：夜间灯光遥感 夜间PM_(2.5)浓度 NPP/VIIRS SVM模型 辐射传输 

分 类 号：X513[环境科学与工程—环境工程]